Abstract
In this paper, a second-order difference scheme is developed to solve two-dimensional two-sided space distributed-order fractional diffusion equation with variable coefficients. In the spatial direction, a second-order difference scheme is proposed, the distribution-order integral is discretized by the Gauss–Legendre quadrature formula and the space fractional derivative is approximated by the weighted and shifted Grünwald–Letnikov operators. In addition, the time direction is discretized into a second-order difference scheme by the Crank–Nicolson method. Therefore, the main numerical scheme is developed. Furthermore, a small perturbation is added to the main difference scheme to construct an alternating-direction implicit scheme. Also, the stability and convergence of the numerical scheme are proved. Finally, some numerical results are provided to show the accuracy and efficiency of the proposed method.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In recent years, distributed-order fractional diffusion equations (DOFDEs) have attracted considerable interest because of its ability to model the processes that become more anomalous in course of time. It is an important tool for modeling ultraslow diffusion [9, 23, 27] or accelerating superdiffusion [8, 36] where a plume of particles spreads at a logarithmic rate. DOFDEs were first proposed in [6]. During these years, the theoretical studies of DOFDEs were carried out by some literatures. For example, in [4], the authors obtained a priori estimation of the solution of the initial boundary value problems of DOFDEs by the maximal principle. In [26], the authors proved the existence of the solution of the boundary value problems of DOFDEs by constructing a formal solution using the Fourier method of variables separation. The authors of [21] discussed the well-posedness of the Cauchy problems of the abstract DOFDEs by functional calculus technique. Ansari et al. [5] used the Mittag-Leffler and Wright functions to obtain the fundamental solution of the DOFDEs with fraction Laplacian in axisymmetric cylindrical configuration. In general, the analytical solutions of many DOFDEs are not easy to gain. Therefore, different numerical methods are worth considering to solve DOFDEs. Zaky et al. [37] developed a spectral tau method based on Legendre polynomials for DOFDEs. In [3], authors studied Galerkin meshless reproducing kernel particle method for neutral delay time-space distributed-order fractional damped diffusion-wave equations. Authors of [19] presented a finite element method for the distributed order time-fractional diffusion equations. Sun et al. [32] proposed a fast and memory saving algorithm for solving distributed-order time-space fractional diffusion equations, and other methods for this type of equations can refer to [14, 16, 38, 42, 43].
As an important class of DOFDEs, the space DOFDEs have attracted extensive attentions, which can be used to simulate an accelerated superdiffusion process [22, 36]. And they mainly include two kinds, one is the Riesz space DOFDEs, the other is two-sided space DOFDEs which is more general. And under certain conditions, the two types of equations can be transformed into each other. For example, authors of [2] studied the following two-dimensional distributed-order Riesz space-fractional diffusion equation
where \(\chi (\beta )\) is non-negative weight function of \(\beta \), and \(\chi (\beta )\) satisfies [1]
\(\frac{\partial ^{\beta } u(x,y,t)}{\partial |x|^{\beta }}\) and \(\frac{\partial ^{\beta } u(x,y,t)}{\partial |y|^{\beta }}\) denote the Riesz space fractional derivative [3, 20]. (1.1) can be written as
let \(\omega (\alpha )=\chi (\alpha +1)\frac{-1}{2\cos \left( \frac{(\alpha +1)\pi }{2}\right) }\), thus (1.1) becomes a two-dimensional two-sided space distributed-order fractional diffusion model [41]. To date there have many numerical methods to be proposed for these equations. Chen et al. [11] derived a fourth-order accurate numerical method for the distributed-order Riesz space fractional diffusion equation. Abbaszadeh et al. [2] presented a fourth-order ADI finite difference scheme to solve the two-dimensional distributed-order Riesz space-fractional diffusion equation. In [36], authors developed a novel finite volume method for the nonlinear two-sided space distributed-order diffusion equation. The general linear method and spectral Galerkin method was proposed by Zhang et al. [41] to solve the nonlinear two-sided space distributed-order diffusion equations.
However, it is worth noting that the diffusion coefficients in (1.2) are constants, and most of the previous work dealt with the problem that the diffusion coefficients are constants, but the diffusion coefficients often depend on the time or space variable in some practical problems [10, 13, 24, 29, 31, 36, 40, 44]. Based on model (1.2), we consider the following two-dimensional two-sided space distributed-order fractional diffusion model with variable diffusivity coefficients:
where \(\Omega =(a,b)\times (c,d)\), \(k_{i,j}(x,y,\alpha )\), \(i,j=1,2\), are nonnegative diffusion coefficients, and they satisfy
f is a given function, and satisfies the Lipschitz condition,
where \(L_{1}\) is a positive constant. Specifically, when \(f(x,y,t,u)=f(x,y,t)\), Eq. (1.3) is a linear problem. And \(\omega (\alpha )\) satisfies
The definitions of operators \(\frac{\partial ^{\alpha } u(x,y,t)}{\partial x^{\alpha }}\), \(\frac{\partial ^{\alpha } u(x,y,t)}{\partial (-x)^{\alpha }}\), \(\frac{\partial ^{\alpha } u(x,y,t)}{\partial y^{\alpha }}\) and \(\frac{\partial ^{\alpha } u(x,y,t)}{\partial (-y)^{\alpha }}\) can be seen [25, 35].
In this paper, we first discretize the distribution-order integral by the Gauss–Legendre quadrature formula, and approximate the space fractional derivative by the weighted and shifted Grünwald–Letnikov operators [15, 34, 35]. Therefore, the second-order accuracy approximation in space can be achieved. In addition, the Crank–Nicolson method [7, 18, 28] are applied to achieve time discretization. Hence, a second-order difference scheme in all variables is developed to solve (1.3). Furthermore, to avoid solving large systems of linear equations, an alternating-direction implicit scheme is constructed by adding a small perturbation to the above second-order difference scheme. Finally, the stability and convergence of the numerical scheme are analyzed.
The outline of this paper is organized as follows. In Sect. 2, we provide the numerical method for solving the two-dimensional two-side space distributed-order fractional diffusion equation with variable diffusivity coefficients. The stability and convergence analysis are proved in Sect. 3. Section 4 presents some numerical results to show the effectiveness of our numerical method. Finally, some conclusions are made in Sect. 5.
2 Numerical method
Let \(\tau =\frac{T}{N}\) be the time step size, and \(h_{1}=\frac{b-a}{M_{1}}\), \(h_{2}=\frac{d-c}{M_{2}}\) be the space step size. And denote \(t_{n}=n\tau \), \(n=0,1,\ldots ,N\), \(x_{i}=a+ih_{1}\), \(i=0,1,\ldots ,M_{1}\), \(y_{j}=c+jh_{2}\), \(j=0,1,\ldots ,M_{2}\). For convenience, we note
If \(\omega (\alpha )\left( A(\alpha ,x,y,t)+B(\alpha ,x,y,t)\right) \in C^{\nu }[0,1]\), using Gauss–Legendre quadrature formula [30, 39] to approximate the integral of Eq. (1.3), we can write (1.3) as
where \(\alpha _{r}=\frac{1+p_{r}}{2}\), and \(p_{r}\), \(w_{r}\), \(r=1,\ldots ,m\), are quadrature points and quadrature weights, respectively. Let
Then, using the Crank–Nicolson method and the weighted and shifted Grünwald–Letnikov operators [25, 33, 35] to approximate Eq. (2.1), it follows that
where
with \(C_r\) is a positive constant. Approximating \(f\left( x_i,y_j,t_{n+\frac{1}{2}},u_{i,j}^{n+\frac{1}{2}}\right) \) with Taylor expansion [36], (2.2) can be converted into
And let
Then, omitting \(r_{ij}^n\), replacing \(u_{ij}^{n}\) with \(U_{ij}^{n}\), Eq. (2.3) can be written as
Further, Eq. (2.4) can be expressed as the following matrix
where
\(H_{1}=\text {diag}(H_{x}(1),H_{x}(2),\ldots ,H_{x}(M_{2}-1))\) with \(H_{x}(k)=(h_{i,j}^{x}(k))_{(M_{1}-1)\times (M_{1}-1)}\), \(1\le k\le M_{2}-1\), \(H_{2}=(h_{i_{1},j_{1}})_{(M_{1}-1)(M_{2}-1)\times (M_{1}-1)(M_{2}-1)}=(h_{i,j})_{(M_{2}-1)\times (M_{2}-1)}\) with \(h_{i,j}=\text {diag}(h_{i,j}^{y}(1),h_{i,j}^{y}(2),\ldots ,h_{i,j}^{y}(M_{1}-1))\), and
To avoid solving large systems of linear equations, we add a small perturbation \(\tau ^{3}(\delta _{x}+\sigma _{x})(\delta _{y}+\sigma _{y})\delta _{t}u_{i,j}^{n+\frac{1}{2}}\) on both sides of (2.3) to construct the following ADI scheme
where
where \(C_{R}\) is a constant. Omitting \(R_{i,j}^{n}\), replacing \(u_{ij}^{n}\) with \(U_{ij}^{n}\), (2.6) can be written as
Further, (2.7) can be expressed as the following matrix
Then, \(U_{i,j}^{n+1}\) can be solved by the following two steps:
where the boundary and initial conditions are
3 Numerical analysis
3.1 Numerical analysis for the difference scheme (2.4)
In this subsection, we first present the following Lemma which are introduced to prove the solvability, convergence and stability of the difference scheme (2.4) will be analyzed and discussed.
Lemma 1
[25, 33, 35] Assume that \(0<\beta <2\), when \(0<\beta <1\), \(\{w_{l}^{\beta }\}\) satisfy
When \(1<\beta <2\), \(\{w_{l}^{\beta }\}\) satisfy
Remark 1
According to Lemma 1, when \(0<\alpha _{r}<1\), \(\sum \nolimits _{l=0}^{M_1-1}w_{l}^{\alpha _{r}}>0\) and \(\sum \nolimits _{l=0,l\ne 1}^{M_1-1}w_{l}^{\alpha _{r}}<0,\ M_1\ge 2\). In addition, \(\sum \nolimits _{l=0}^{M_1-1}w_{l}^{\alpha _{r}+1}<0\) and \(\sum \nolimits _{l=0,l\ne 1}^{M_1-1}w_{l}^{\alpha _{r}+1}>0,\ M_1\ge 3\).
Theorem 1
Suppose that \(0<\alpha _{r}<1\), \(M_{1},M_{2}\ge 3\), \(k_{i,j}(x,y,\alpha )\), \(i,j=1,2\), satisfy condition (1.4), then the difference scheme (2.4) is uniquely solvable.
Proof
In order to prove the unique solvability of the difference scheme (2.4), we need to prove \(I+H\) is strictly diagonally dominant. For all \(i=1,\ldots ,M_{1}-1\), \(k=1,2,\ldots ,M_{2}-1\), by means of Remark 1, we have
Thus, \(H_{x}(k)\) is strictly diagonally dominant, furthermore, \(H_{1}\) is strictly diagonally dominant. Similarly, \(h_{i,i}^{y}(k)>\sum \nolimits _{j=1,j\ne i}^{M_{2}-1}|h_{i,j}^{y}(k)|\), \(H_{2}\) is strictly diagonally dominant. Therefore, \(I+H\) is strictly diagonally dominant. In other words, the matrices \(I+H\) is invertible, which guarantees the difference scheme (2.4) is uniquely solvable. \(\square \)
Lemma 2
Suppose that \(0<\alpha _{r}<1\), \(M_{1},M_{2}\ge 3\), \(k_{i,j}(x,y,\alpha )\), \(i,j=1,2\), satisfy condition (1.4), then the eigenvalues of \(I+H\) are all greater than 1.
Proof
Based on the Greschgorin’s theorem [12, 17], the eigenvalues \(\{\lambda _{i}\}\) of matrix \(H_{x}(k)\) satisfy
then using Theorem 1, it follows that
namely, the eigenvalues of the matrix \(H_{x}(k)\) are all greater than 0. Furthermore, since \(H_{1}\) is a block diagonal matrix, the eigenvalues of \(H_{1}\) are also all greater than 0. Similarly, it can be verified that the eigenvalues of \(H_{2}\) are all greater than 0. Let \(\lambda _{1}\) and \(\lambda _{2}\) be any eigenvalues of \(H_{1}\) and \(H_{2}\), respectively, then there are \(\lambda _{1}>0\) and \(\lambda _{2}>0\), thus the eigenvalues of \(I+H\) are all greater than 1, \(\square \)
Now, suppose that (1.3) has a unique sufficient smooth solution \(u\in C_{x,y,t}^{5,5,3}\left( \Omega \times [0,T]\right) \), let \(U_{i,j}^{n}\) and \(\widetilde{U}_{i,j}^{n}\) be the solution and numerical solution of the difference scheme (2.4), respectively. Define \(\epsilon _{i,j}^{n}=U_{i,j}^{n}-\widetilde{U}_{i,j}^{n}\), substitution into (2.4) leads into
And (3.4) can be written as
where
with
Let \(\lambda _{i,j}\) be any eigenvalues of H, then the eigenvalues of \(I+H\) and \(I-H\) are \(1+\lambda _{i,j}\) and \(1-\lambda _{i,j}\), respectively. From Lemma 2, we know \(\lambda _{i,j}>0\), thus \(1+\lambda _{i,j}>1\). And let \(\left( 1+\lambda _{p,q}\right) \) be the eigenvalue corresponding to row \((M_2-1)(q-1)+p\) of matrix \(I+H\), then \(\left( 1-\lambda _{p,q}\right) \) be the eigenvalue corresponding to row \((M_2-1)(q-1)+p\) of matrix \(I-H\). Further, we have
Then, we present the following Theorem based on Eq. (3.6) to illustrate the stability of the difference scheme (2.4).
Theorem 2
Suppose that \(0<\alpha <1\), \(k_{i,j}(x,y,\alpha )\), \(i,j=1,2\), satisfy condition (1.4), then the difference scheme (2.4) is unconditionally stable.
Proof
Denote \(||\epsilon ^{n+1}||_{\infty }=|\epsilon _{p,q}^{n+1}|=\max \nolimits _{1\le i\le M_{1}-1,1\le j\le M_{2}-1}|\epsilon _{i,j}^{n+1}|\), it follows from (3.5) and (3.6) that
using the Lipschitz condition (1.5), we have
where \(C_1\) is a positive constant, further,
namely,
Summing up for n from 1 to k, we have
applying the Gronwall inequality [12] to above inequality yield
This completes the proof. \(\square \)
Now, we consider the convergence of the difference scheme (2.4). Let \(u_{ij}^{n}\) and \(U_{i,j}^{n}\) be the exact solution and numerical solution of (1.3), respectively. Define \(e_{i,j}^{n}=U_{i,j}^{n}-u_{i,j}^{n}\) and \(e^{n}=(e_{1,1}^{n},e_{2,1}^{n},\ldots ,e_{M_{1}-1,1}^{n},\ldots ,e_{1,M_{2}-1}^{n},\ldots ,e_{M_{1}-1,M_{2}-1}^{n})^{T}\). From (2.3) and (2.4), we get
Then, we present the following Theorem based on Eq. (3.6) to illustrate the convergence of the difference scheme (2.4).
Theorem 3
Suppose that \(0<\alpha \le 1\) and \(\tau \) is small enough, \(k_{i,j}(x,y,\alpha )\), \(i,j=1,2\), satisfy condition (1.4), then the error of the difference scheme (2.4) satisfy
where \(C_{1}\) and \(C_r\) are positive constants.
Proof
Denote \(||E^{n+1}||_{\infty }=|e_{p,q}^{n+1}|=\max \nolimits _{1\le i\le m_{1}-1,1\le j\le m_{2}-1}|e_{i,j}^{n+1}|\), it follows (3.13) that
where \(r^{n}=(r_{1,1}^{n},r_{2,1}^{n},\ldots ,r_{M_{1}-1,1}^{n},\ldots ,r_{1,M_{2}-1}^{n},\ldots ,r_{M_{1}-1,M_{2}-1}^{n})^{T}\). Using the Lipschitz condition (1.5), we have
where \(C_1\) is a positive constant, further,
namely,
Summing up for n from 1 to k, and notice \(\left\| e^{0}\right\| _{\infty }=0\), we have
applying the Gronwall inequality [12] to above inequality yield
This completes the proof. \(\square \)
3.2 Numerical analysis for the difference scheme (2.7)
In this subsection, the solvability, convergence and stability of the difference scheme (2.7) will be analyzed and discussed.
Theorem 4
Suppose that \(0<\alpha _{r}<1\), \(k_{i,j}(x,y,\alpha )\), \(i,j=1,2\), satisfy condition (1.4), then the difference scheme (2.7) is uniquely solvable.
Proof
In order to prove the unique solvability of the difference scheme (2.7), we need to prove \(I+H_{1}\) and \(I+H_{2}\) are strictly diagonally dominant. By means of Theorem 1, we know \(H_{1}\) and \(H_{2}\) are strictly diagonally dominant. Therefore, \(I+H_{1}\) and \(I+H_{2}\) are strictly diagonally dominant. In other words, the matrices \(I+H_{1}\) and \(I+H_{2}\) are invertible, which guarantees the difference scheme (2.7) is uniquely solvable. \(\square \)
Lemma 3
Suppose that \(0<\alpha _{r}<1\), \(M_{1},M_{2}\ge 3\), \(k_{i,j}(x,y,\alpha )\), \(i,j=1,2\), satisfy condition (1.4), then the eigenvalues of \(I+H_{1}\) and \(I+H_{2}\) are all greater than 1.
Proof
Using Lemma 2, we know the eigenvalues of \(H_{1}\) and \(H_{2}\) are all greater than 0. Thus, the eigenvalues of \(I+H_{1}\) and \(I+H_{2}\) are all greater than 1, \(\square \)
Suppose that (1.3) has a unique sufficient smooth solution \(u\in C_{x,y,t}^{5,5,3}\left( \Omega \times [0,T]\right) \), let \(U_{i,j}^{n}\) and \(\widetilde{U}_{i,j}^{n}\) be the solution and numerical solution of the difference scheme (2.7), respectively. Define \(\epsilon _{i,j}^{n}=U_{i,j}^{n}-\widetilde{U}_{i,j}^{n}\), substitution into (2.7) leads into
Further, (3.22) can be expressed as the following matrix
where
and
Let \(\lambda _{i,j}\) and \(\mu _{i,j}\) be any eigenvalues of \(I+H_{1}\) and \(I+H_{2}\), respectively, then the eigenvalues of W and K are \((1+\lambda _{i,j})(1+\mu _{i,j})\) and \((1-\lambda _{i,j})(1-\mu _{i,j})\), respectively. From Lemma 3, we know \(1+\lambda _{i,j}>1\) and \(1+\mu _{i,j}>1\). And let \(\left( 1+\lambda _{p,q}\right) \left( 1+\mu _{p,q}\right) \) be the eigenvalue corresponding to row \((M_2-1)(q-1)+p\) of matrix W, then \(\left( 1-\lambda _{p,q}\right) \left( 1-\mu _{p,q}\right) \) be the eigenvalue corresponding to row \((M_2-1)(q-1)+p\) of matrix K. Further, we have
and
Then, we present the following Theorem based on Eqs. (3.24), (3.25) to illustrate the stability of the difference scheme (2.7).
Theorem 5
Suppose that \(0<\alpha <1\), \(k_{i,j}(x,y,\alpha )\), \(i,j=1,2\), satisfy condition (1.4), then the difference scheme (2.7) is unconditionally stable.
Proof
Denote \(||\epsilon ^{n+1}||_{\infty }=|\epsilon _{p,q}^{n+1}|=\max \limits _{1\le i\le M_{1}-1,1\le j\le M_{2}-1}|\epsilon _{i,j}^{n+1}|\), it follows from (3.23), (3.24) and (3.25) that
Using the Lipschitz condition (1.5) yiled
where \(C_{1}\) is a positive constant. Using Lemma 3, it follows that
namely,
Summing up for n from 1 to k, we have
applying the Gronwall inequality [12] to above inequality yield
This completes the proof. \(\square \)
Now, we consider the convergence of the difference scheme (2.7). Let \(u_{ij}^{n}\) and \(U_{i,j}^{n}\) be the exact solution and numerical solution of (1.3), respectively. Define \(e_{i,j}^{n}=U_{i,j}^{n}-u_{i,j}^{n}\) and \(e^{n}=(e_{1,1}^{n},e_{2,1}^{n},\ldots ,e_{M_{1}-1,1}^{n},\ldots ,e_{1,M_{2}-1}^{n},\ldots ,e_{M_{1}-1,M_{2}-1}^{n})^{T}\). From (2.3) and (2.7), we get
Then, we present the following Theorem based on Eqs. (3.24), (3.25) to illustrate the stability of the difference scheme (2.7).
Theorem 6
Suppose that \(0<\alpha \le 1\) and \(\tau \) is small enough, \(k_{i,j}(x,y,\alpha )\), \(i,j=1,2\), satisfy condition (1.4), then the error of the difference scheme (2.7) satisfy
where \(C_{1}\) and \(C_R\) are positive constants.
Proof
Denote \(||E^{n+1}||_{\infty }=|e_{p,q}^{n+1}|=\max \nolimits _{1\le i\le m_{1}-1,1\le j\le m_{2}-1}|e_{i,j}^{n+1}|\), it follows from (3.32), (3.24) and (3.25) that
where \(R^{n}=(R_{1,1}^{n},R_{2,1}^{n},\ldots ,R_{M_{1}-1,1}^{n},\ldots ,R_{1,M_{2}-1}^{n},\ldots ,R_{M_{1}-1,M_{2}-1}^{n})^{T}\). Using Lemma 3, it follows that
namely,
Summing up for n from 1 to k, and notice \(\left\| e^{0}\right\| _{\infty }=0\), we have
applying the Gronwall inequality [12] to above inequality yield
This completes the proof. \(\square \)
4 Numerical examples
In this section, we give three examples to demonstrate the accuracy and efficiency of the numerical schemes (2.4) and (2.7). All numerical results are carried out by Matlab R2018a software, and the computation time (CPU) is measured in seconds. The errors are given by
And in Example 1 and Example 2, the spatial direction and the temporal direction convergence orders are given by
In Examples 1–3, when \(\tau =h_{1}=h_{2}\), the convergence orders are given by
Example 1. Consider the model (1.3) with \(T=2\), \(\Omega =(0,1)\times (0,1)\), \(\omega (\alpha )=2\cos (\frac{\pi \alpha }{2})\), \(k_{11}(x,y,\alpha )=k_{21}(x,y,\alpha )=\frac{3-e^{xy\alpha }}{2}\), \(k_{12}(x,y,\alpha )=k_{22}(x,y,\alpha )=\frac{3+e^{xy\alpha }}{2}\), \(f(x,y,t,u)=f(x,y,t)\), f(x, y, t) and \(\varphi (x,y)\) are determined by the exact solution \(u(x,y,t)=e^{-t}x^{2}(1-x)^{2}y^{2}(1-y)^{2}\).
This is a linear numerical example. The errors and the convergence orders of the numerical schemes (2.4) and (2.7) in the spatial and temporal directions are shown in Tables 1 and 2, respectively. As can be seen from Tables 1 and 2 that the convergence orders are closed to theoretical results. And as can be seen from the numerical results, numerical scheme (2.7) has a slightly higher error than numerical scheme (2.4), but the CPU time of numerical scheme (2.7) is lower than that of numerical scheme (2.4), especially when the number of nodes is larger, which indicates that numerical scheme (2.7) is more suitable for dealing with large and sparse systems of linear equations. Figure 1 give the maximum errors of the numerical schemes (2.4) and (2.7) for solving Example 1 in distributed order. From it, we can observe that the error hardly changes anymore when the number of quadrature nodes increases to a certain number, this suggests that errors in the numerical scheme at this point are dominated by errors in space and time. Therefore, in the calculation process, we compute distribution-order integral using Gauss–Legendre quadrature rule with \(m=4\). Table 3 gives the maximum errors and convergence orders of numerical scheme (2.7) based on different \(\alpha \) when the number of spatial nodes is the same as the number of temporal nodes, in which case the model (1.3) can be reduced to two-sided space fractional diffusion equations with variable coefficients [13]. Figures 2 and 3 present the approximation solution at \(\tau =\frac{1}{40}\), \(m=4\) as can be seen from the Figs. 2 and 3, the numerical solution is in well accordance with the exact solution, it’s verified the validity of the two numerical schemes.
Example 2. Consider the model (1.3) with \(T=1\), \(\Omega =(0,1)\times (0,1)\), \(\omega (\alpha )=-2\Gamma (4-\alpha )\cos \left( \frac{\pi (\alpha +1)}{2}\right) \), \(k_{11}(x,y,\alpha )=k_{21}(x,y,\alpha )=k_{12}(x,y,\alpha )=k_{22}(x,y,\alpha )=\frac{-1}{2\cos \left( \frac{\pi (\alpha +1)}{2}\right) }\), \(f(x,y,t,u)=f(x,y,t)+g(u)\), f(x, y, t, u) and \(\varphi (x,y)\) are determined by the exact solution \(u(x,y,t)=t^{3}x^{4}(1-x)^{4}y^{4}(1-y)^{4}\). This example can be reduced to model (1.1) with \(\chi (\beta )=-2\Gamma (5-\beta )\cos (\frac{\pi \beta }{2})\). When \(g(u)=0\), this is a linear numerical example. The errors and the convergence orders of the method in [2] and the numerical scheme (2.7) in the spatial and temporal directions are shown in Tables 4 and 5, respectively. From Table 4, it can be seen that when the number of spatial nodes is small, the numerical scheme (2.7) has less error than the method in [2], but when the number of nodes is large, the method in [2] is more advantageous than the numerical scheme (2.7) and the convergence order is higher than that of the numerical scheme (2.7). As can be seen from Table 5, the numerical scheme (2.7) has a lower error than the method in [2], but the CPU time is more than the method in [2]. When \(g(u)=-u(1+u)\), this is a nonlinear numerical example. And assume \(\tau =h_{1}=h_{2}\), \(m=4\) in this example. The errors and the convergence orders of the numerical scheme (2.4) and the numerical scheme (2.7) are shown in Table 6. The numerical results show that the convergence order is consistent with the theoretical value, verifying the accuracy of the numerical scheme. And Fig. 4a gives the max absolute errors of the numerical schemes (2.4) and (2.7) for different values of \(\tau \).
Example 3. Consider the model (1.3) with \(T=1.5\), \(\Omega =(0,1)\times (0,1)\), \(\omega (\alpha )=2e^{\alpha }\), \(k_{11}(x,y,\alpha )=\frac{1-x^{2}y\alpha }{2}\), \(k_{21}(x,y,\alpha )=\frac{1-xy^{2}\alpha }{2}\), \(k_{12}(x,y,\alpha )=\frac{1+x^{2}y\alpha }{2}\), \(k_{22}(x,y,\alpha )=\frac{1+xy^{2}\alpha }{2}\), \(f(x,y,t,u)=g(u)\), \(g(u)=u(1+u)\), \(\varphi (x,y)=\sin (1)x^{3}(1-x)^{3}y^{3}(1-y)^{3}\).
This is a nonlinear numerical example without exact solution, thus, we take the approximate solutions with \(m=4\), \(N= 1000\) and \(M_1=M_2= 100\) as reference solutions in this example. And assume \(\tau =h_{1}=h_{2}\), \(m=4\) in this example. Table 7 give some numerical results when different values of \(\tau \) are taken. From Table 7, it can be concluded that the convergence orders of both two numerical schemes is close to 2, and the error of numerical scheme (2.4) is lower than that of numerical scheme (2.7), which is consistent with the theoretical analysis. Moreover, the CPU time of the numerical scheme (2.7) is lower than that of the numerical scheme (2.4). Especially when the number of nodes is large. This indicates that the numerical scheme (2.7) is better suited to handle large systems of sparse linear equations. Figure 4b gives the max absolute errors of the numerical schemes (2.4) and (2.7) for different values of \(\tau \).
5 Conclusion
In this paper, we developed a second order in both space and time numerical scheme for two-dimensional two-sided space distributed-order fractional diffusion equation with variable coefficients. In addition, a small perturbation is added to this numerical scheme to construct an alternating-direction implicit scheme. Subsequently, we proved that the difference scheme is unconditionally stable and convergent with the accuracy of \(O(m^{-v}+h_{1}^{2}+h_{2}^{2}+\tau ^{2})\). Finally, some numerical results are given to show the stability and convergence of our numerical scheme. And the numerical results indicates that the numerical scheme (2.7) is better suited to handle large systems of sparse linear equations. However, the model in this paper is too restrictive on the diffusion coefficient, and the computational cost of the numerical scheme is too high. In future work, we will work on developing higher-order numerical methods or developing fast iterative algorithms to solve models with less restriction on diffusion coefficients.
Data availibility
Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.
References
Abbaszadeh, M.: Error estimate of second-order finite difference scheme for solving the Riesz space distributed-order diffusion equation. Appl. Math. Lett. 88, 179–185 (2019)
Abbaszadeh, M., Dehghan, M.: A POD-based reduced-order Crank–Nicolson/fourth-order alternating direction implicit (ADI) finite difference scheme for solving the two-dimensional distributed-order Riesz space-fractional diffusion equation. Appl. Numer. Math. 158, 271–291 (2020)
Abbaszadeh, M., Dehghan, M.: A Galerkin meshless reproducing kernel particle method for numerical solution of neutral delay time-space distributed-order fractional damped diffusion-wave equation. Appl. Numer. Math. 169, 44–63 (2021)
Al-Refa, M., Luchko, Y.: Analysis of fractional diffusion equations of distributed order: maximum principles and their applications. Analysis 36(2), 123–133 (2016)
Ansari, A., Derakhshan, M., Askari, H.: Distributed order fractional diffusion equation with fractional Laplacian in axisymmetric cylindrical configuration. Commun. Nonlinear. Sci. 113, 106590 (2022)
Caputo, M.: Mean fractional-order-derivatives differential equations and filters. Ann. Univ. Ferrara 41, 73–84 (1995)
Çelik, C., Duman, M.: Crank–Nicolson method for the fractional diffusion equation with the Riesz fractional derivative. J. Comput. Phys. 231, 1743–1750 (2012)
Chechkin, A., Gorenflo, R., Sokolov, I.: Retarding subdiffusion and accelerating superdiffusion governed by distributed-order fractional diffusion equations. Phys. Rev. E 66, 046129 (2002)
Chechkin, A., Gorenflo, R., Sokolov, I., Gonchar, V.: Distributed order time fractional diffusion equation. Fract. Calc. Appl. Anal. 6, 259–279 (2003)
Chen, S., Liu, F., Jiang, X., Turner, I., Anh, V.: A fast semi-implicit difference method for a nonlinear two-sided space-fractional diffusion equation with variable diffusivity coefficients. Appl. Math. Comput. 257, 591–601 (2015)
Chen, X., Chen, J., Liu, F., Sun, Z.: A fourth-order accurate numerical method for the distributed-order Riesz space fractional diffusion equation. Numer. Methods Partial Differ. Equ. 39, 1266–1286 (2023)
Cheng, X., Duan, J., Li, D.: A novel compact ADI scheme for two-dimensional Riesz space fractional nonlinear reaction–diffusion equations. Appl. Math. Comput. 346, 452–464 (2019)
Feng, L., Zhuang, P., Liu, F., Turner, I.: A fast second-order accurate method for a two-sided space-fractional diffusion equation with variable coefficients. Comput. Math. Appl. 73, 1155–1171 (2017)
Gao, G., Sun, Z.: Two alternating direction implicit difference schemes for two-dimensional distributed-order fractional diffusion equations. J. Sci. Comput. 66, 1281–1312 (2016)
Gao, G., Sun, Z.: Two difference schemes for solving the one-dimensional time distributed-order fractional wave equations. Numer. Algorithm 74, 675–697 (2017)
Gao, X., Yin, B., Li, H., Liu, Y.: TT-M FE method for a 2D nonlinear time distributed-order and space fractional diffusion equation. Math. Comput. Simul. 181, 117–137 (2021)
Golub, G., Loan, C.: Matrix Computations. Johns Hopkins University Press, Baltimore (1996)
Hou, T., Tang, T., Yang, J.: Numerical analysis of fully discretized Crank–Nicolson scheme for fractional-in-space Allen–Cahn equations. J. Sci. Comput. 72, 1–18 (2017)
Huang, C., Chen, H., An, N.: \(\beta \)-robust superconvergent analysis of a finite element method for the distributed order time-fractional diffusion equation. J. Sci. Comput. 90(1), 44 (2022)
Huang, J., Zhang, J., Arshad, S., Tang, Y.: A numerical method for two-dimensional multi-term time-space fractional nonlinear diffusion-wave equations. Appl. Numer. Math. 159, 159–173 (2021)
Jia, J., Peng, J., Li, K.: Well-posedness of abstract distributed-order fractional diffusion equations. Commun. Pure Appl. Anal. 13, 605–621 (2014)
Jia, J., Wang, H.: Analysis of a hidden memory variably distributed-order space-fractional diffusion equation. Appl. Math. Lett. 124, 107617 (2022)
Kochubei, A.: Distributed order calculus and equations of ultraslow diffusion. J. Math. Anal. Appl. 340, 252–281 (2008)
Liu, F., Anh, V., Turner, I.: Numerical solution of the space fractional Fokker–Planck equation. J. Comput. Appl. Math. 166, 209–219 (2004)
Liu, F., Zhuang, P., Liu, Q.: Numerical Methods of Fractional Partial Differential Equations and Their Applications. Beijing Science Press, Beijing (2015)
Luchko, Y.: Boundary value problems for the generalized time-fractional diffusion equation of distributed order. Fract. Calc. Appl. Anal. 12, 409–422 (2009)
Meerschaert, M., Scheffler, H.: Stochastic model for ultraslow diffusion. Stoch. Process. Appl. 116, 1215–1235 (2006)
Ran, M., Zhang, C.: A conservative difference scheme for solving the strongly coupled nonlinear fractional Schrödinger equations. Commun. Nonlinear Sci. Numer. 41, 64–83 (2016)
Schumer, R., Benson, D., Meerschaert, M., Baeumer, B.: Fractal mobile/immobile solute transport. Water Resour. Res. 39, 1296 (2003)
Shen, J., Tang, T., Wang, L.: Spectral Methods. Springer Series in Computational Mathematics. Springer, Heidelberg (2011)
Su, N.: The fractional Boussinesq equation of groundwater flow and its applications. J. Hydrol. 547, 403–412 (2017)
Sun, L., Fang, Z., Lei, S., Sun, H., Zhang, J.: A fast algorithm for two-dimensional distributed-order time-space fractional diffusion equations. Appl. Math. Comput. 425, 127095 (2022)
Tian, W., Zhou, H., Deng, W.: A class of second order difference approximations for solving space fractional diffusion equations. Math. Comput. 84, 1703–1727 (2015)
Wang, Z., Vong, S.: Compact difference schemes for the modified anomalous fractional sub-diffusion equation and the fractional diffusion-wave equation. J. Comput. Phys. 277, 1–15 (2014)
Yang, S., Liu, F., Feng, L., Turner, I.: Efficient numerical methods for the nonlinear two-sided space-fractional diffusion equation with variable coefficients. Appl. Numer. Math. 157, 55–68 (2020)
Yang, S., Liu, F., Feng, L., Turner, I.: A novel finite volume method for the nonlinear two-sided space distributed-order diffusion equation with variable coefficients. J. Comput. Appl. Math. 388, 113337 (2021)
Zaky, M., Machado, J.: Multi-dimensional spectral tau methods for distributed-order fractional diffusion equations. Comput. Math. Appl. 79, 476–488 (2020)
Zhang, H., Liu, F., Jiang, X., Turner, I.: Spectral method for the two-dimensional time distributed-order diffusion-wave equation on a semi-infinite domain. J. Comput. Appl. Math. 399, 113712 (2022)
Zhang, H., Liu, F., Jiang, X., Zeng, F., Turner, I.: A Crank–Nicolson ADI Galerkin–Legendre spectral method for the two-dimensional Riesz space distributed-order advection-diffusion equation. Comput. Math. Appl. 76, 2460–2476 (2018)
Zhang, Y., Benson, D., Meerschaert, M., LaBolle, E.: Space-fractional advection–dispersion equations with variable parameters: diverse formulas, numerical solutions, and application to the macrodispersion experiment site data. Water Resour. Res. 43, W05439 (2007)
Zhang, Y., Fan, Y., Li, Y.: General linear and spectral Galerkin methods for the nonlinear two-sided space distributed-order diffusion equation. Comput. Math. Appl. 113, 1–12 (2022)
Zhao, J., Zhang, Y., Xu, Y.: Implicit Runge–Kutta and spectral Galerkin methods for Riesz space fractional/distributed-order diffusion equation. Comput. Appl. Math. 39, 47 (2020)
Zhao, J., Zhang, Y., Xu, Y.: Implicit Runge–Kutta and spectral Galerkin methods for the two-dimensional nonlinear Riesz space distributed-order diffusion equation. Appl. Numer. Math. 157, 223–235 (2020)
Zhao, X., Xu, Q.: Efficient numerical schemes for fractional sub-diffusion equation with the spatially variable coefficient. Appl. Math. Model. 38, 3848–3859 (2014)
Acknowledgements
The authors thank the reviewers for their comments, which have significantly improved the presentation.
Funding
The authors would like to acknowledge the support provided by the National Natural Science Foundation of China (12101089) and the Natural Science Foundation of Sichuan Province (2022NSFSC1844).
Ethics declarations
Conflict of interest
The authors declare no competing interests.
Ethical approval
Not applicable.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Wang, Y., Huang, J. & Li, H. A second-order difference scheme for two-dimensional two-sided space distributed-order fractional diffusion equations with variable coefficients. J. Appl. Math. Comput. 70, 1–26 (2024). https://doi.org/10.1007/s12190-023-01954-9
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s12190-023-01954-9
Keywords
- Two-dimensional distributed-order fractional diffusion equation
- Alternating direction implicit method
- Variable coefficient
- Stability and convergence